Page 410 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

2.2. Refine and merge lines
 Refine crop lines
The lines that remain from the first phase are discrete, and it is not possible to achieve the final crop lines by connecting
them. An algorithm is needed to achieve more regular and direct lines. At the first step for this goal a binary mask made by
considering the greater thickness of the lines. The lines are then obtained with a centrifugal skeletonize filterBy applying the
Hough Transform algorithm to the skeleton image, more direct crop lines are obtained. The outputs of refine step was shown in
Figure 4.

Fig 4: process for refine initial lines. (A) Show lines by specific width. (B) Apply skeletonization operator. (C) Detect stright
lines using Hough Transform.
 Merge lines using a series of rules
In this step, the proposed methodology creates a line scanner with the declination equals to the detected crop row
orientation (α), and bias (b) is shown in Figure 5. In Figure 6 each of the peaks and valleys represents either a crop row or a soil
row, while the slopes between the peaks and the valleys are the translation between the crop and soil rows. Through averaging
the peaks and the valley's values, the crop rows are detected by comparing these values. The bias of the different planting lines
can be a criterion for classifying the lines. This graph can help for a tile and a set of classified lines according to the same specific
bias.
− 1
2
= (4)
− 1
2

= + 1 (5)
1

Fig 5: Calculate bias for a particular line

5